Grounded Docs MCP Server
Give your AI coding assistant up-to-date, version-specific documentation to reduce hallucinations and keep context fresh.
- Source repo
- arabold/docs-mcp-server
- Stars
- ★ 1.8k
- Last updated
- today
- License
- MIT
- Primary language
- TypeScript
- FA score
- 50/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformChatGPT · Claude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A developer using Cursor or VS Code's Cline extension wants AI assistance that is grounded in the exact library version used in the project (e.g., React 18 vs 19).
- Not a fit if
- Users stuck on Node.js versions below 22
- Those wanting zero-config setup without embedding models or Playwright
- Source review
- 50/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
Grounded Docs MCP Server is a self-hosted documentation indexing service that fetches official docs from websites, GitHub, npm, PyPI, and local files, providing an always-current context for AI coding assistants. It supports a wide range of formats including PDF, Word, Markdown, source code, and more, and offers both a web UI and CLI. It integrates with MCP-compatible clients like Claude and Cline, supports optional semantic search via embedding models, and runs entirely locally for privacy. It is positioned as an open-source alternative to Context7, Nia, and Ref.Tools. The server exposes an SSE-based MCP endpoint on port 6280 by default.
The tool performs the following operations: scrapes documentation pages from specified URLs and stores them in an index (e.g., npx @arabold/docs-mcp-server@latest scrape react https://react.dev/reference/react); performs keyword or semantic search on the index (search react "useEffect cleanup"); fetches a single page and converts it to Markdown (fetch-url); provides an MCP server endpoint (default port 6280) for AI clients to query; imports documentation from local folders and ZIP archives; and optionally configures embedding models such as OpenAI, Ollama, Gemini, and Azure to improve search quality.
- A developer using Cursor or VS Code's Cline extension wants AI assistance that is grounded in the exact library version used in the project (e.g., React 18 vs 19).
- A team with strict data privacy requirements needs a fully local documentation retrieval service that never sends code to the cloud.
- A technical writer needs to quickly convert official documentation (e.g., TypeScript handbook) into Markdown for an internal knowledge base.
- An AI application developer wants to integrate version-aware documentation querying into a custom MCP client, reducing hallucinations.
- A DevOps engineer wants to containerize a documentation indexing service for development environments, ensuring the latest versions are tracked.
How do you install or deploy this agent?
Requires Node.js 22+. Run directly via npx: npx @arabold/docs-mcp-server@latest. Alternatively, use Docker: docker run --rm -v docs-mcp-data:/data -v docs-mcp-config:/config -p 6280:6280 ghcr.io/arabold/docs-mcp-server:latest --protocol http --host 0.0.0.0 --port 6280. After starting, open http://localhost:6280 to add documentation.
How do you use this agent?
- Index docs:
npx @arabold/docs-mcp-server@latest scrape react https://react.dev/reference/react. 2. Search:npx @arabold/docs-mcp-server@latest search react "useEffect cleanup" --output yaml. 3. Fetch a page:npx @arabold/docs-mcp-server@latest fetch-url https://react.dev/reference/react/useEffect. 4. Configure MCP client (e.g., Claude Desktop) by adding to config:{"mcpServers": {"docs-mcp-server": {"type": "sse", "url": "http://localhost:6280/sse"}}}. 5. Optional: setOPENAI_API_KEYto enable embedding models.
What are this agent's strengths and limitations?
- Fetches documentation directly from official sources on demand, ensuring freshness and reducing hallucinations
- Queries target the exact library versions in your project, improving accuracy
- Runs entirely on your machine, so your code and data never leave your network
- Compatible with any MCP client and supports Docker deployment for easy setup
- Supports a broad range of file formats including PDF, Office, and 90+ source code languages
- Requires Node.js 22+, potentially excluding older environments
- Optional embedding models add configuration overhead (e.g., OpenAI, Ollama) and may incur costs
- Large documentation indexes may consume significant local storage and computation
- Hash-routed SPAs require Playwright, adding an extra dependency
- When using external embedding providers like OpenAI, network access and API costs become a consideration
How does this agent compare with similar options?
As an open-source alternative to Context7, Nia, and Ref.Tools, this tool emphasizes fully local operation and version-specific accuracy, whereas Context7 and similar tools may store indexes in the cloud or offer a broader documentation library.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Grounded Docs MCP Server This agent | 50 · Major gaps | CLIFree + model costs | ★ 1.8k | today | TypeScript | ChatGPT · Claude Code |
| Uni-CLI | 72 · Some gaps | CLIFree | ★ 235 | 8d ago | TypeScript | ChatGPT · Codex · Claude Code |
| Microsoft Learn MCP Server | 43 · Major gaps | MCP serverFree | ★ 2k | 1mo ago | TypeScript | ChatGPT · Codex · Claude Code · Claude.ai |
| wigolo | 54 · Major gaps | CLIFree + model costs | ★ 5.5k | 22d ago | TypeScript | Codex · Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: the project promises local, private operation (README) and supports optional authentication (OAuth2/OIDC) and network access controls (security.md). However, no explicit least-privilege design (e.g., requesting permissions on demand) or user confirmation mechanisms (e.g., confirmation for dangerous operations) were found. Data flow transparency: README and docs describe the data flow (scrape, index, query) but do not detail how data is processed, stored, and transmitted. Sensitive data handling: API keys (e.g., OPENAI_API_KEY) are supported but secure storage is not explained. Dependency security: many dependencies are used, but no dependency audit or vulnerability scanning evidence is provided. External effects: scraping external sites may generate network requests, but rate limiting or respect for target sites is not mentioned. Rollback: no rollback mechanism for indexes or configurations is mentioned. Source attribution: documentation sources are explicit (official docs), but how source integrity is verified is not explained. Deductions: lack of least-privilege, user confirmation, rollback, and other key security features.
Evidence shows: the project has tests (unit, integration, E2E) and CI, but no test coverage or stability data is provided. Dependency availability: many dependencies, but no dependency locking or mirroring strategy is provided. Failure messages: CLI tests verify error output, but comprehensive error handling documentation is not provided. Deductions: dependency availability is not fully ensured, and failure message coverage is limited.
Evidence shows: README and docs cover multiple usage scenarios (CLI, MCP, Docker) and provide configuration options (e.g., embedding models, authentication). Capability boundaries: docs list supported formats and sources, but do not specify limits (e.g., max pages, concurrency). Trigger precision: CLI commands and MCP tools are clearly defined, but detailed parameter descriptions are not provided. Environment fit: multiple deployment modes (local, Docker, distributed) are supported, but detailed configuration for all environments is not provided. Deductions: capability boundaries and trigger precision are not described in enough detail.
Evidence shows: clear information architecture (README, docs directory), detailed installation notes (npx, Docker), stable naming (package name, commands), examples and FAQ exist (examples in README), known limitations partially mentioned (e.g., hash routing), clear license (MIT), versioning exists (package.json version), maintenance responsibility is not explicit (no maintainer or contribution guide specified). Deductions: known limitations are not comprehensive, and maintenance responsibility is unclear.
Evidence shows: multiple output formats (JSON, YAML, Markdown) for easy integration. Marginal value: as an alternative to Context7, provides local, private documentation indexing. Cost-benefit: free and open source, but requires Node.js 22+ and possibly API keys (embedding models). Deductions: no performance benchmarks or comparisons with other tools are provided.
Evidence shows: claims in README (e.g., supported formats) are supported by docs, but no independent verification is provided. Cross-source corroboration: no third-party verification or user reviews are provided. Fact-inference separation: README distinguishes feature descriptions and recommendations, but does not explicitly mark which are inferences. Deductions: lack of independent verification and explicit fact/inference separation.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- The project depends on many third-party libraries, but no dependency audit or vulnerability scanning evidence is provided; conduct a security review before use.
- The project supports scraping external websites, which may generate network requests; configure network access controls and respect target sites' robots.txt.
- The project supports API keys (e.g., OPENAI_API_KEY); ensure keys are stored securely and not committed to version control.
- The project does not provide a rollback mechanism; consider regular backups of index data.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Breaking changes - Scraping an existing library version now requires explicit replacement: use CLI --replace or MCP/API replace: true, use refresh to update in place, or use CLI --no-clean to append. Upgrade remote workers before or alongside their coordinators. ([#517](https://github.com/arabold/docs-mcp-server/issues/517), [#515](https://github.com/arabold/docs-mcp-server/issues/515), [#168](https://github.com/arabold/docs-mcp-server/issues/168), [#436](https://github.com/arabold/docs-mcp-server/issues/436)) …View on GitHub ↗ -
Release notes excerpt
[3.2.1](https://github.com/arabold/docs-mcp-server/compare/v3.2.0...v3.2.1) (2026-09-27) Bug Fixes * web: keep add header working when the portal is served over plain http ([3d45d67](https://github.com/arabold/docs-mcp-server/commit/3d45d672984229997e328eeb35b3e1b4c71428b8)), closes [#514](https://github.com/arabold/docs-mcp-server/issues/514)View on GitHub ↗ -
Release notes excerpt
[3.2.0](https://github.com/arabold/docs-mcp-server/compare/v3.1.0...v3.2.0) (2026-09-21) Bug Fixes * embeddings: address PR review comments on spec/code inconsistencies ([7beddd3](https://github.com/arabold/docs-mcp-server/commit/7beddd31f5b956f4feceed4a710b66dfbaf885f2)) * embeddings: match provider prefixes case-insensitively ([ceb51ef](https://github.com/arabold/docs-mcp-server/commit/ceb51efab16742d0a669a0a515234b69e0040866)) …View on GitHub ↗